ebook img

Filling the Gap between Metric Regularity and Fixed Points: The Linear Openness of Compositions PDF

0.18 MB·English
Save to my drive
Quick download
Download
Most books are stored in the elastic cloud where traffic is expensive. For this reason, we have a limit on daily download.

Preview Filling the Gap between Metric Regularity and Fixed Points: The Linear Openness of Compositions

FILLING THE GAP BETWEEN METRIC REGULARITY AND FIXED POINTS: THE LINEAR OPENNESS OF COMPOSITIONS 1 M. DUREA 1 Faculty of Mathematics, ”Al. I. Cuza” University, 0 Bd. Carol I, nr. 11, 700506– Ia¸si, Romania, 2 e-mail: [email protected] n a J R. STRUGARIU 1 Department of Mathematics, ”Gh. Asachi” Technical University, 3 Bd. Carol I, nr. 11, 700506– Ia¸si, Romania, ] e-mail: [email protected] A F . h t Abstract: This paper is devoted to the investigation of an important issue recently brought into attention a m by a recentpaper of Arutyunov: the relationbetween openness of compositionof set-valued maps and fixed [ point results. More precisely, we prove a general result concerning the openness of compositions and then we show that this result covers and implies most of the known openness results. In particular, we reobtain 1 v several recent results in this field, including a fixed point theorem of Dontchev and Frankowska. 9 5 Keywords: compositionofset-valuedmappings·linearopenness·metricregularity·Lipschitz-likeproperty 9 5 · implicit multifunctions · fixed points . 1 0 Mathematics Subject Classification (2010): 90C30 · 49J53 · 54C60 1 1 : v 1 Introduction i X r The equivalent properties of metric regularity and openness at linear rate, which are nowadays a mainly studied in the context of the set-valued maps, have an important history. Their origins are in the open mapping principle for linear operators obtained in the 1930s by Banach and Schauder. Subsequently, a nonlinear extension of open mapping principle was obtained by Lyusternik (1934) and Graves (1950) and, besides the importance on this result, another important contribution is its proof which became, along the years, a powerful instrument in the effort of getting generalizations, extensionsandabetterunderstandingoftheseseminalworks. Anotherlandmarkwastheextension of research to the case of set-valued maps with closed and convex graph and was done by Ursescu in 1975 and Robinson in 1976, respectively. The famous Robinson-Ursescu Theorem and a series of works of Milyutin in 1970s concerning the preservation of regularity (and openness) of set- valued maps under functional perturbation give a strong impetus to this direction of research and, therefore, since 1980s to our time many mathematicians have participated into a joint effort on the development and understanding of these problems. We mention here only a few, having major 1 contributions to the field: J. P. Aubin, A. Dontchev, H. Frankowska, A. Ioffe, B. S. Mordukhovich, J. P. Penot, R. T. Rockafellar, S. M. Robinson, C. Ursescu. We remark that in several works (of J. P. Penot and C. Ursescu, for instance) dealing with openness results, the techniques based on the original proof of Lyusternik-Graves Theorem (which is in fact an iteration procedure) have been replaced by the use of Ekeland Variational Principle. Detailed accounts on the historical facts, as well as on the evolution of terminology are given in the monographs of Rockafellar and Wets [23], Mordukhovich [18] and Dontchev and Rockafellar [8]. The present paper is motivated by some (very) recent developments, which suggest that the studyofmetricregularity canbedoneinrelationwithsomefixedpointsresults. Thisnewapproach is suggested in the monograph of Dontchev and Rockafellar [8], and it is more precisely emphasized in the work of Arutyunov [3]. Subsequently, the relation between fixed points results and metric regularity are further investigated in [6] and [16]. In [6], [16], some compositions of set-valued maps are involved and, in fact, the degrees of generality of considered compositions represent, in some extent, the main novelties in both these papers. More precisely, Ioffe [16] investigates separately coincidence fixed point results and openness of compositions, while Dontchev and Frankowska in [6] present a fixed point theorem which gen- eralized the previous result of Arutyunov, and, moreover, show that their result implies several well-known results concerning the metric regularity of sum of set-valued maps. Moreover, these papers emphasize the fact that the link between fixed points and metric regularity of the involved multifunctions has a purely metric behavior. As a consequence, their proofs make use of well-fitted adaptations oftheinitialiterative procedureof Lyusternik-Graves type,andworkongeneralmetric spaces, under local completeness assumptions. In this paper we propose ourselves to bring into light a global approach and to fill several implications whicharenotpreviouslygiven. Havingthisaiminmind,wefollow thenextprocedure. Firstly, we prove some implicit multifunctions assertions which, besides their own importance are usedseveraltimesasauxiliaryresults. Secondly,weproveageneralopennessresultforcompositions ofset-valuedmaps. OurproofisarefinementofthemethodbasedonEkelandVariationalPrinciple. In particular, we prove that from our result one can get a nontrivial extension of the corresponding result of Ioffe. Moreover, the same result implies as well Dontchev-Frankowska fixed point theorem and this implication (i.e. from openness assertions to fixed points results) is not given elsewhere (uptoourknowledge). Atthispoint, itisclear thatalltheimplications fromDontchev-Frankowska fixed point theorem apply, and therefore our openness result of compositions implies several very important results in this topic. At every step of our investigation we are trying to give an accurate account on the usual difficulties that arise at certain points of the proofs and to interrelate our results and arguments to the ones in literature. The setting we assume in this work is that of Banach spaces. Note that most of the results work equally well on normed vector spaces, with corresponding completeness assumptions around the reference points. The paper is organized as follows. In the second section we introduce the basic notations, concepts and we recall some known results. The third section contains the main results of the paper. We present here some implicit multifunction theorems which are important as separate results, but their conclusions are in force in the proof of other results. Next, we formulate and prove the main result of the paper which is an openness theorem for compositions of set-valued maps. Then, on this basis, one result of Ioffe is significantly extended and completed. The last section is devoted to the proof of Dontchev-Frankowska fixed point theorem as a consequence of our main result and several interpretations and possibilities of generalizations are presented. The paper ends with a short conclusion section. 2 2 Preliminaries This section contains some basic definitions and results used in the sequel. In what follows, we suppose that all the involved spaces are Banach. In this setting, B(x,r) and D(x,r) denote the open and the closed ball with center x and radius r, respectively. Sometimes we write D for the X closed unit ball of X. If x ∈ X and A ⊂ X, one defines the distance from x to A as d(x,A) := inf{kx−ak| a ∈ A}. As usual, we use the convention d(x,∅) = ∞. For a non-empty set A ⊂ X we put clA for its topological closure. When we work on a product space, we consider the sum norm, unless otherwise stated. Consider now a multifunction F : X ⇉ Y. The domain and the graph of F are denoted respectively by DomF := {x ∈ X |F(x) 6= ∅} and GrF = {(x,y) ∈ X ×Y |y ∈ F(x)}. If A ⊂ X then F(A) := [F(x). The inverse set-valued map of F is F−1 : Y ⇉ X given by x∈A F−1(y) = {x ∈ X | y ∈ F(x)}. Recall that a multifunction F is inner semicontinuous at (x,y) ∈ GrF if for every open set D ⊂ Y with y ∈ D, there exists a neighborhoodU ∈ V(x) such that for every x′ ∈ U, F(x′)∩D 6= ∅ (where V(x) stands for the system of the neighborhoods of x). We remind now the concepts of openness at linear rate, metric regularity and Lipschitz-likeness of a multifunction around the reference point. Definition 2.1 Let L > 0, F :X ⇉ Y be a multifunction and (x,y)∈ GrF. (i) F is said to be open at linear rate L > 0, or L−open around (x,y) if there exist a positive number ε > 0 and two neighborhoods U ∈ V(x), V ∈ V(y) such that, for every ρ∈ (0,ε) and every (x,y) ∈ GrF ∩[U ×V], B(y,ρL) ⊂ F(B(x,ρ)). (2.1) The supremum of L > 0 over all the combinations (L,U,V,ε) for which (2.1) holds is denoted by lopF(x,y) and is called the exact linear openness bound, or the exact covering bound of F around (x,y). (ii) F is said to be Lipschitz-like, or has Aubin property around (x,y) with constant L > 0 if there exist two neighborhoods U ∈ V(x), V ∈ V(y) such that, for every x,u∈ U, F(x)∩V ⊂ F(u)+Lkx−ukD . (2.2) Y The infimum of L > 0 over all the combinations (L,U,V) for which (2.2) holds is denoted by lipF(x,y) and is called the exact Lipschitz bound of F around (x,y). (iii) F is said to be metrically regular around (x,y) with constant L > 0 if there exist two neighborhoods U ∈V(x), V ∈V(y) such that, for every (x,y) ∈ U ×V, d(x,F−1(y)) ≤ Ld(y,F(x)). (2.3) The infimum of L > 0 over all the combinations (L,U,V) for which (2.3) holds is denoted by regF(x,y) and is called the exact regularity bound of F around (x,y). The next proposition contains the well-known links between the notions presented above. For more details about the proof, see [18, Theorems 1.49, 1.52]. 3 Proposition 2.2 Let F : X ⇉ Y be a multifunction and (x,y) ∈ GrF. Then F is open at linear rate around (x,y) iff F−1 is Lipschitz-like around (y,x) iff F is metrically regular around (x,y). Moreover, in every of the previous situations, (lopF(x,y))−1 = lipF−1(y,x) = regF(x,y). It is well known that the corresponding ”at point” properties are significantly different from the ”around point” ones. Let us introduce now some of these notions. For more related concepts we refer to [1]. Definition 2.3 Let L > 0, F :X ⇉ Y be a multifunction and (x,y)∈ GrF. (i) F is said to be open at linear rate L, or L−open at (x,y) if there exists a positive number ε > 0 such that, for every ρ∈ (0,ε), B(y,ρL) ⊂ F(B(x,ρ)). (2.4) The supremum of L > 0 over all the combinations (L,ε) for which (2.4) holds is denoted by plopF(x,y) and is called the exact punctual linear openness bound of F at (x,y). (ii) F is said to be pseudocalm with constant L, or L−pseudocalm at (x,y), if there exists a neighborhood U ∈ V(x) such that, for every x ∈ U, d(y,F(x)) ≤ Lkx−xk. (2.5) The infimum of L > 0 over all the combinations (L,U) for which (2.5) holds is denoted by psdclmF(x,y) and is called the exact bound of pseudocalmness for F at (x,y). (iii) F is said to be metrically hemiregular with constant L, or L−metrically hemiregular at (x,y) if there exists a neighborhood V ∈ V(y) such that, for every y ∈ V, d(x,F−1(y)) ≤ Lky−yk. (2.6) The infimum of L > 0 over all the combinations (L,V) for which (2.6) holds is denoted by hemregF(x,y) and is called the exact hemiregularity bound of F at (x,y). For more details about these concepts, see [2], [12]. The next proposition presents the corre- sponding equivalences between the ”at point” notions introduced before. Proposition 2.4 Let L > 0, F :X ⇉ Y and (x,y)∈ GrF. Then F is L−open at (x,y) iff F−1 is L−1−pseudocalm at (y,x) iff F is L−1−metrically hemiregular at (x,y). Moreover, in every of the previous situations, (plopF(x,y))−1 = psdclmF−1(y,x) = hemregF(x,y). RecallthatL(X,Y)denotesthenormedvectorspaceoflinearboundedoperatorsactingbetween X and Y. If A ∈ L(X,Y), then the ”at” and ”around point” notions do coincide. In fact, A is metrically regular around every x ∈ X iff A is metrically hemiregular at every x ∈ X iff A is open with linear rate around every x ∈ X iff A is open with linear rate at every x ∈ X iff A is surjective. Moreover, in every of these cases we have hemregA= regA = (plopA)−1 = (lopA)−1 = (A∗)−1 , (cid:13) (cid:13) (cid:13) (cid:13) 4 where A∗ ∈ L(Y∗,X∗) denotes the adjoint operator and hemregA, regA, plopA and lopA are common for all the points x ∈ X (see, for more details, [2, Proposition 5.2]). Finally, weintroducethecorrespondingpartial notions oflinear openness,metric regularity and Lipschitz-like property around the reference point for a parametric set-valued map. Definition 2.5 Let L > 0, F : X ×P ⇉ Y be a multifunction, ((x,p),y) ∈ GrF and for every p ∈ P, denote F (·) := F(·,p). p (i) F is said to be open at linear rate L > 0, or L−open, with respect to x uniformly in p around ((x,p),y) if there exist a positive number ε > 0 and some neighborhoods U ∈ V(x), V ∈ V(p), W ∈ V(y) such that, for every ρ∈ (0,ε), every p ∈ V and every (x,y) ∈ GrF ∩[U ×W], p B(y,ρL)⊂ F (B(x,ρ)), (2.7) p The supremum of L > 0 over all the combinations (L,U,V,W,ε) for which (2.7) holds isdenoted by lop F((x,p),y) and is called the exact linear openness bound, or the exact covering bound of F x c in x around ((x,p),y). (ii) F is said to be Lipschitz-like, or has Aubin property, with respect to x uniformly in p around ((x,p),y) with constant L > 0 if there exist some neighborhoods U ∈ V(x), V ∈ V(p), W ∈ V(y) such that, for every x,u ∈ U and every p ∈ V, F (x)∩W ⊂ F (u)+Lkx−ukD . (2.8) p p Y The infimum of L > 0 over all the combinations (L,U,V,W) for which (2.8) holds is denoted by lip F((x,p),y) and is called the exact Lipschitz bound of F in x around ((x,p),y). x c (iii) F is said to be metrically regular with respect to x uniformly in p around ((x,p),y) with constant L > 0 if there exist some neighborhoods U ∈ V(x), V ∈ V(p), W ∈ V(y) such that, for every (x,p,y) ∈ U ×V ×W, d(x,F−1(y)) ≤ Ld(y,F (x)). (2.9) p p The infimum of L > 0 over all the combinations (L,U,V,W) for which (2.9) holds is denoted by reg F((x,p),y) and is called the exact regularity bound of F in x around ((x,p),y). x c Similarly, one can define the notions of linear openness, metric regularity and Lipschitz-like property with respect to p uniformly in x, and the corresponding exact bounds. 3 Linear openness of compositions We start the main section of the paper with a refinement of a result previously given in [12]. Here, we present some conclusions in a slightly different form but without inner semicontinuity assumptions (see as well the comments after the proof). Theorem 3.1 Let X,P be metric spaces, Y be a normed vector space, H : X × P ⇉ Y be a set-valued map and (x,p,0) ∈ GrH. Denote by H (·) := H(·,p), H (·) := H(x,·). p x (i) If H is open with linear rate c > 0 with respect to x uniformly in p around (x,p,0), then there exist α,β,γ > 0 such that, for every (x,p) ∈ B(x,α)×B(p,β), d(x,S(p)) ≤ c−1d(0,H(x,p)∩B(0,γ)). (3.1) 5 Suppose, in addition, that Y is a normed vector space and H is Lipschitz-like with respect to p uniformly in x around (x,p,0). Then S is Lipschitz-like around (p,x) and lipS(p,x)≤ c−1lip H((x,p),0). (3.2) p c (ii) If H is open with linear rate c > 0 with respect to p uniformly in x around (x,p,0), then there exist α,β,γ > 0 such that, for every (x,p) ∈ B(x,α)×B(p,β), d(p,S−1(x)) ≤ c−1d(0,H(x,p)∩B(0,γ)). (3.3) If, moreover, H is Lipschitz-like with respect to x uniformly in p around (x,p,0), then S is metrically regular around (p,x) and regS(p,x) ≤ c−1lip H((x,p),0). (3.4) x c Proof. We will prove only the first item, because for the second one it suffices to observe that, defining the multifunction T := S−1, the proof is completely symmetrical, using T instead of S. Moreover, using Proposition 2.2, we know that regS(p,x) = lipT(x,p) and then (3.4) follows from (3.2). For the (i) item, we know that there exist r,s,t,c,ε > 0 such that, for every ρ ∈ (0,ε), every p ∈ B(p,t) and every (x,y) ∈ GrH ∩[B(x,r)×B(0,s)], p B(y,cρ) ⊂ H (B(x,ρ)). p Take now ρ ∈ (0,min{ε,c−1s}). Set α := r, β := t, γ := cρ and fix arbitrary (x,p) ∈ B(x,α)× B(p,β). If H(x,p) ∩ B(0,cρ) = ∅, then d(0,H(x,p) ∩ B(0,γ)) = +∞ and (3.1) trivially holds. Suppose next that H(x,p)∩B(0,cρ) 6= ∅. If 0 ∈ H(x,p), then 0 ∈ H(x,p)∩B(0,cρ), and, again, (3.1) trivially holds. Suppose now that 0 6∈ H(x,p)∩B(0,cρ). Then for every ξ > 0, there exists y ∈H(x,p)∩B(0,cρ) such that ξ ky k < d(0,H(x,p)∩B(0,cρ))+ξ. ξ Because d(0,H(x,p)∩B(0,cρ)) < cρ, we can choose ξ sufficiently small such that d(0,H(x,p)∩ B(0,cρ))+ξ < cρ. Consequently, 0 ∈ B(y ,d(0,H(x,p)∩B(0,cρ))+ξ)⊂ B(y ,cρ). (3.5) ξ ξ Observe now that x ∈ B(x,r), p ∈ B(p,t), y ∈ B(0,d(0,H(x,p)∩B(0,cρ))+ξ) ⊂ B(0,cρ) ⊂ ξ B(0,s), y ∈ H(x,p) and denote ρ := c−1(d(0,H(x,p)∩B(0,cρ))+ξ)< ρ < ε. ξ 0 But we know that B(y ,cρ ) ⊂ H (B(x,ρ )), ξ 0 p 0 hence, using also (3.5), one obtains that there exists x ∈ B(x,ρ ) such that 0∈ H(x ,p), which is 0 0 0 equivalent to x ∈S(p). Then 0 d(x,S(p)) ≤ d(x,x )< ρ = c−1(d(0,H(x,p)∩B(0,cρ))+ξ). 0 0 Making ξ → 0, we obtain (3.1). Suppose now that H is Lipschitz-like with respect to p uniformly in x around (x,p,0). Then there exist l,a,b,τ > 0 such that τ < cρ and for every x ∈ B(x,a) and every p ,p ∈ B(p,b), 1 2 H(x,p )∩D(0,τ) ⊂ H(x,p )+ld(p ,p )D . (3.6) 1 2 1 2 Y 6 Take α := min{a,α}, β = min{b,β,(2l)−1τ}, p ,p ∈ B(p,β) and x ∈ S(p )∩B(x,α). Then 1 2 1 0 ∈ H(x,p )∩D(0,τ), whence, using (3.6), there exists y′′ ∈ D such that y′ := l·d(p ,p )y′′ ∈ 1 Y 1 2 H(x,p ) with ky′k ≤ l·[d(p ,p)+d(p,p )] ≤ τ < cρ. Hence, y′ ∈ H(x,p )∩B(0,γ), so using (3.1), 2 1 2 2 we get that d(x,S(p )) ≤ c−1d(0,H(x,p )∩B(0,γ)) ≤ c−1 y′ ≤ c−1ld(p ,p ). 2 2 (cid:13) (cid:13) 1 2 (cid:13) (cid:13) Consequently, because l can be chosen arbitrarily close to lip H((x,p),0)), it follows that S is p c Lipschitz-like around (p,x) and lipS(p,x)≤ c−1lip H((x,p),0)). The proof is now complete. (cid:3) p c Remark 3.2 If, in addition to the assumptions of Theorem 3.1, H is inner semicontinuous at (x,p,0), then the relation (3.1) becomes d(x,S(p)) ≤ c−1d(0,H(x,p)). Also, as one can see from the precedent proof, P can be taken to be just topological space (see, for more details, [11, Theorem 3.2]). We present the next result as a by-product of Theorem 3.1. Note that different versions of the second part of the following lemma are done in [2, Theorem 3.5] (with functions instead of multifunctions) and in [16, Lemma 2]. Here we also obtain some extra conclusions for a general situation. Taking into account that this lemma will be used in the sequel, we prefer to give all the details of its proof. Lemma 3.3 LetY,Z,W benormedvectorspaces, G :Y×Z ⇉ W beamultifunctionand(y,z,w)∈ Y ×Z ×W be such that w ∈ G(y,z). Consider next the implicit multifunction Γ : Z ×W ⇉ Y defined by Γ(z,w) := {y ∈ Y | w ∈ G(y,z)}. Suppose that the following conditions are satisfied: (i) G is Lipschitz-like with respect to z uniformly in y around ((y,z),w) with constant D ≥ 0; (ii) G is open at linear rate with respect to y uniformly in z around ((y,z),w) with constant C > 0. Then the multifunction H :Y ×Z ×W ⇉ W, given by H(y,(z,w)) := G(y,z)−w for every (y,z,w) ∈ Y ×Z ×W, is Lipschitz-like with respect to (z,w) uniformly in y around (y,(z,w),0). Moreover, there exists γ > 0 such that for every (z,w),(z′,w′) ∈ D(z,γ) × D(w,γ), and for every δ > 0, 1+δ Γ(z,w)∩D(y,γ) ⊂ Γ(z′,w′)+ (D z−z′ + w−w′ )D . (3.7) (cid:13) (cid:13) (cid:13) (cid:13) Y C (cid:13) (cid:13) (cid:13) (cid:13) Proof. Define P := Z ×W and then H :Y ×P ⇉ W. Observe also that Γ(z,w) = {y ∈ Y | 0 ∈ H(y,(z,w))}, and denote H (·) := H(·,(z,w)). Because for every z close to z we know from (ii) that G is (z,w) z C−open at points from its graph around (y,w), we can conclude that there exists ε > 0 such that for every ρ ∈(0,ε), every z ∈B(z,ε) and every (y,w′) ∈ GrG ∩[B(y,ε)×B(w,ε)], z ′ B(w ,Cρ)⊂ G (B(y,ρ)). z 7 Hence,forevery(z,w) ∈ B(z,ε)×B(w,2−1ε)andevery(y,u) ∈GrH ∩[B(y,ε)×B(0,2−1ε)], (z,w) setting w′ := u+w ∈ B(w,ε), we get that B(u,C−1ρ) = B(w′,Cρ)−w ⊂ G (B(y,ρ))−w = H (B(y,ρ)). z (z,w) But this shows, applying Theorem 3.1, that there exist β > 0 such that, for every (y,z,w) ∈ B(y,β)×B(z,β)×B(w,β), d(y,Γ(z,w)) ≤ C−1d(0,H(y,(z,w))∩B(0,β)). (3.8) We want to prove that there exists c > 0 such that (D + 1)c < β, for every y ∈ B(y,c), (z,w),(z′,w′) ∈ B(z,c)×B(w,c), H(y,(z,w))∩D(0,c) ⊂ H(y,(z′,w′))+(D z−z′ + w−w′ )D . (3.9) (cid:13) (cid:13) (cid:13) (cid:13) Z (cid:13) (cid:13) (cid:13) (cid:13) In particular, we will prove the first part of the conclusion. Because of (i), we know that there exists a > 0 such that for every y ∈ B(y,a) and every z,z′ ∈ B(z,a), G(y,z)∩D(w,a) ⊂ G(y,z′)+D z−z′ D . (3.10) (cid:13) (cid:13) Z (cid:13) (cid:13) Choose now c ∈ (0,min{2−1a,(D + 1)−1β}) and take arbitrary y ∈ B(y,c), (z,w),(z′,w′) ∈ B(z,c)×B(w,c). Furthermore, choose u′ ∈ H(y,(z,w))∩D(0,c). Then ku′+w−wk ≤ 2c < a, whence u′+w ∈ G(y,z)∩D(w,a) and because of (3.10), one obtains successively that u′+w ∈ G(y,z′)+D z−z′ D , (cid:13) (cid:13) Z u′ ∈ G(y,z′)−w′(cid:13)+(w−(cid:13)w′)+D z−z′ D , (cid:13) (cid:13) Z u′ ∈ H(y,(z′,w′))+(D z−z′ +(cid:13) w−w(cid:13)′ )D . (cid:13) (cid:13) (cid:13) (cid:13) Z (cid:13) (cid:13) (cid:13) (cid:13) Take now γ ∈ (0,min{β,2−1c}), (z,w),(z′,w′) ∈ B(z,γ)×B(w,γ) and y ∈ Γ(z,w)∩D(y,γ). Hence, 0 ∈ H(y,(z,w)) ∩D(0,c). Then, using (3.9), we have that there exists η ∈ D such that Z (Dkz−z′k+kw−w′k)η ∈ H(y,(z′,w′))∩B(0,β) (because (D+1)c < β). Finally, using (3.8), we get that d(y,Γ(z′,w′)) ≤ C−1d(0,H(y,(z′,w′))∩B(0,β)) ≤ C−1(D z−z′ + w−w′ ), (cid:13) (cid:13) (cid:13) (cid:13) (cid:13) (cid:13) (cid:13) (cid:13) which completes the proof. (cid:3) We are now inposition to formulate andto prove ourmain result, an opennessresultfora fairly general set-valued composition. Theorem 3.4 Let X,Y,Z,W be Banach spaces, F :X ⇉ Y, F :X ⇉ Z and G :Y ×Z ⇉ W be 1 2 three multifunctions and (x,y,z,w) ∈ X ×Y ×Z ×W such that (x,y) ∈GrF , (x,z)∈ GrF and 1 2 ((y,z),w) ∈ GrG. Let H : X ⇉ W be given by H(x) := G(F (x),F (x)) for every x ∈ X 1 2 and suppose that the following assumptions are satisfied: (i) GrF , GrF and GrG are locally closed around (x,y), (x,z) and ((y,z),w); 1 2 (ii) F is open at linear rate L > 0 around (x,y); 1 8 (iii) F is Lipschitz-like around (x,z) with constant M > 0; 2 (iv) G is open at linear rate with respect to y uniformly in z around ((y,z),w) with constant C > 0; (v) G is Lipschitz-like with respect to z uniformly in y around ((y,z),w) with constant D ≥ 0; (vi) LC −MD > 0. Then there exists ε> 0 such that, for every ρ ∈(0,ε), B(w,(LC −MD)ρ) ⊂H(B(x,ρ)). Moreover, for every ρ∈ (0,2−1ε) and every (x,y,z,w) ∈ B(x,2−1ε)×B(y,2−1ε)×B(z,2−1ε)× B(w,2−1ε) such that (y,z) ∈ F (x)×F (x) and w ∈ G(y,z), 1 2 B(w,(LC −MD)ρ) ⊂H(B(x,ρ)). Proof. Using (i), one can find α > 0 such that GrF ∩cl[B(x,α)×B(y,α)], GrF ∩cl[B(x,α)× 1 2 B(z,α)] and GrG∩cl[B(y,α)×B(z,α)×B(w,α)] are closed. Also, using (ii) and (iii), there exist β > 0 such that, for every (x′,y′) ∈ GrF ∩[B(x,β)×B(y,β)], F is L−open at (x′,y′) and for 1 1 every (v′,u′) ∈ GrF−1∩[B(z,β)×B(x,β)], F−1 is M−1−open at (v′,u′). Finally, using (iv) and 2 2 (v), we can find γ > 0 such that for every (z,w),(z′,w′) ∈ B(z,γ)×B(w,γ) and every δ > 0 1+δ Γ(z,w)∩B(y,γ) ⊂Γ(z′,w′)+ (D z−z′ + w−w′ )B . (3.11) (cid:13) (cid:13) (cid:13) (cid:13) Y C (cid:13) (cid:13) (cid:13) (cid:13) Without loosing the generality, using (v), we can suppose also that for every y ∈ B(y,γ) and every z,z′ ∈ B(z,γ), G(y,z)∩B(w,γ) ⊂ G(y,z′)+D z−z′ D . (3.12) (cid:13) (cid:13) Y (cid:13) (cid:13) Fix ε := min{α,L−1α,M−1α,(LC + MD)−1α,β,2−1L−1β,2−1M−1β,γ,2−1L−1γ,2−1M−1γ, 2−1(LC +MD)−1γ} and take ρ∈ (0,ε). Define the multifunction (F ,F ) : X ⇉ Y ×Z by (F ,F )(x) := F (x)×F (x) and observe 1 2 1 2 1 2 that (x,y,z) ∈ Gr(F ,F ). Because of the choice of ε, we have that the set Ω∩clA is closed, where 1 2 A := B(x,ρ)×B(y,Lρ)×B(z,Mρ)×B(w,(LC +MD)ρ), and (3.13) Ω := {(x,y,z,w) ∈ X ×Y ×Z ×W |(y,z) ∈ (F ,F )(x) and w ∈ G(y,z)}. (3.14) 1 2 Take u∈ B(w,(LC−MD)ρ). We must prove that u∈ H(B(x,ρ)). There exists τ ∈ (0,1) such that ku−wk < τ(LC −MD)ρ. Endow the space X ×Y ×Z ×W with the norm k(p,q,r,s)k := τ(LC −MD)max{kpk,L−1kqk,M−1krk,(LC +MD)−1ksk} 0 and apply the Ekeland variational principle to the function h :Ω∩clA→ R , + h(p,q,r,s) := ku−sk. Then one can find a point (a,b,c,d) ∈ Ω∩clA such that ku−dk ≤ ku−wk−k(a,b,c,d)−(x,y,z,w)k (3.15) 0 and ku−dk ≤ ku−wk+k(a,b,c,d)−(p,q,r,s)k , ∀(p,q,r,s) ∈ Ω∩clA. (3.16) 0 9 From (3.15) we have that τ(LC −MD)max{ka−xk,L−1kb−yk,M−1kc−zk,(LC +MD)−1kd−wk} = k(a,b,c,d)−(x,y,z,w)k ≤ ku−wk < τ(LC −MD)ρ, 0 hence (a,b,c,d) ∈ A, and, in particular, a ∈ B(x,ρ). If u= d, then u ∈H(a) ⊂ H(B(x,ρ)) and the desired assertion is proved. We want to show that u = d is the sole possible situation. For this, suppose by means of contradiction that u 6= d. Fix ω > 0 such that LC −ω > 0 and define next v := (LC −ω)ku−dk−1(u−d). Take arbitrary ζ ∈ (0,2−1ρ). Remark that, from the choice of ε, b ∈ B(y,γ),c ∈ B(z,γ) and d,d+ζv ∈ B(w,γ). Hence, using (3.11) for δ ∈ (0,(LC −ω)−1ω), we get that b ∈ Γ(c,d)∩B(y,γ) ⊂ Γ(c,d+ζv)+(L−ε′)ζD , Y where ε′ := C−1(ω−δ(LC−ω)) ∈ (0,L). Therefore, there exists q ∈ D such that b+(L−ε′)ζq ∈ Y Γ(c,d+ζv), or, equivalently, d+ζv ∈ G(b+(L−ε′)ζq,c). Now, from the L− openness of F at (a,b) ∈ GrF ∩[B(x,β)×B(y,β)], there exists ε < 2−1ρ 1 1 0 such that, for every ζ ∈ (0,ε ), 0 ′ b+(L−ε)ζq ∈ B(b,Lζ) ⊂ F (B(a,ζ)). (3.17) 1 Consequently, there exists p with kpk< 1 such that b+(L−ε′)ζq ∈ F (a+ζp). 1 Also, using the M−1− openness of F−1 at (c,a) ∈ GrF−1 ∩[B(z,β)×B(x,β)], we can find 2 2 ε < ε such that for every ζ ∈ (0,ε ), 1 0 1 B(a,ζ)⊂ F−1(B(c,Mζ)). (3.18) 2 Hence, one can find e ∈ B(c,Mζ) such that a+ζp ∈ F−1(e) or, equivalently, e ∈ F (a+ζp). 2 2 Becausewecanwritee= c+Mζrwithkrk< 1,wefinallyhavethat(a+ζp,b+(L−ε′)ζq,c+Mζr)∈ Gr(F ,F ). 1 2 From the choice of ε, we know that c,c+Mζr ∈ B(z,γ) and b+(L−ε′)ζq ∈ B(y,γ). Using now (3.12), we get that for every ζ ∈(0,ε ), 1 ′ d+ζv ∈ G(b+(L−ε)ζq,c)∩B(w,γ) ⊂ G(b+(L−ε′)ζq,c+Mζr)+MDζD , W so there exist s ∈ D such that d+ζv+MDζs∈ G(b+(L−ε′)ζq,c+Mζr). W In conclusion, for every ζ ∈ (0,ε ), one can find (p,q,r,s) ∈ (B ,D ,B ,D ) such that 1 X Y Z W (a+ζp,b+(L−ε′)ζq,c+Mζr,d+ζv+MDζs)∈ Ω∩A.We use now (3.16) to obtain that ′ ku−dk ≤ ku−(d+ζv+MDζs)k+ζ (p,(L−ε)q,Mr,v +MDs) (3.19) (cid:13) (cid:13) 0 (cid:13) (cid:13) ′ ≤ ku−d−ζvk+MDζ +ζ (p,(L−ε)q,Mr,v +MDs) . (cid:13) (cid:13) 0 (cid:13) (cid:13) But ku−d−ζvk= |ku−dk−ζ(LC −ω)|. 10

See more

The list of books you might like

Most books are stored in the elastic cloud where traffic is expensive. For this reason, we have a limit on daily download.